Oracle Database 23ai AI Vector Search — Is It Free, Is It an Option, What Does It Cost?

Oracle 23ai AI Vector Search licensing is one of the cleanest commercial constructs Oracle has shipped in five years — and one of the most aggressively positioned by the Oracle account team. The VECTOR data type, vector indexes (HNSW and IVF), and the SQL similarity operators are bundled with Oracle Database 23ai Enterprise Edition at no additional charge. There is no separately licensed AI Vector Search option in the database options price list. Customers running 23ai Enterprise Edition can deploy vector workloads inside the existing Enterprise Edition licence footprint.

That sentence is the marketing one-liner. The buyer-side analysis has to go three steps further — what edition rules apply, what the workload footprint does to the licence position, and what Oracle's playbook around the supporting feature stack looks like when LMS arrives. For the broader Oracle Database licensing context, the Oracle Database licensing master guide sets out the option model and the Core Factor Table mechanics. This piece works the AI Vector Search line item the way an Oracle insider preparing a 23ai upgrade business case would work it.

The Oracle 23ai AI Vector Search licensing model

Bundled with Enterprise Edition — not a separate option

Oracle Database 23ai Enterprise Edition includes AI Vector Search at no additional charge. The VECTOR data type, the HNSW and IVF vector indexes, the vector similarity SQL operators (VECTOR_DISTANCE, VECTOR_DIMENSION_COUNT and the related primitives), and the SQL Macro patterns for retrieval-augmented generation are all part of the Enterprise Edition feature set in 23ai. Customers do not require a separate option licence in the way that Oracle Partitioning or Advanced Compression require separate option licences.

Standard Edition 2 — explicitly excluded

Standard Edition 2 customers cannot use AI Vector Search in 23ai. Oracle has scoped the VECTOR data type and the vector indexing capabilities to Enterprise Edition only. The licensing implication is significant — SE2 customers running mid-market deployments who want vector search functionality face a binary decision: upgrade the database to Enterprise Edition (with the Named User Plus or Processor Metric step-up that implies), or use an external vector store (pgvector on PostgreSQL, Pinecone, Weaviate, Qdrant, or similar) and integrate at the application tier. Oracle's account teams will push the Enterprise Edition upgrade route; the buyer-side analysis has to model both paths.

Autonomous Database 23ai — bundled in consumption pricing

Oracle Autonomous Database 23ai includes AI Vector Search in the consumption-based ECPU pricing model — no separate uplift, no separate option line. The economic exposure on Autonomous shifts from a per-Processor licence position to the ECPU-hour consumption profile, which typically increases materially when vector embeddings are deployed at scale. The licence model is simpler; the consumption forecast becomes the critical buyer-side artefact.

The total cost of an Oracle 23ai AI Vector Search deployment

The base Enterprise Edition position

The Enterprise Edition list prices above are indicative against the 2026 Oracle Technology Global Price List. The AI Vector Search bundling means that an existing Enterprise Edition deployment running 23ai picks up vector search capability with no incremental option cost. The economic question is not the bundle — it is what the vector workload does to the supporting licence position.

The supporting option attack surface

A production AI Vector Search deployment at enterprise scale typically generates three supporting option dependencies. Partitioning becomes the right approach for large vector embedding tables — vector data is typically high-cardinality, write-heavy on ingest, and partition-friendly on retrieval. Advanced Compression provides vector compression that materially reduces storage footprint for embedding columns. Active Data Guard enables real-time replication of vector data to a standby for disaster recovery — a pattern most production deployments end up requiring. Each of these is a separately licensed Enterprise Edition option carrying its own per-Processor list price. The bundle covers the feature; the option attach covers the production deployment.

The 19c-to-23ai upgrade trap

Why customers want to upgrade to 23ai

The AI Vector Search functionality is the headline reason customers want to upgrade from Oracle Database 19c to 23ai — the SQL-native vector search capability, the integration with Oracle Generative AI Service, the Select AI function for natural-language queries, and the RAG pipeline patterns that enterprise AI deployments now require. The upgrade narrative is compelling — the existing Oracle Database investment extends to cover the AI workload without the application architecture having to integrate a separate vector store.

The licensing implication of the upgrade

The upgrade from 19c to 23ai does not in itself trigger an incremental licence position for customers already on Enterprise Edition with current support. The 23ai release is included under the existing Software Update Licence and Support entitlement. The buyer-side defence is to verify three things forensically before the upgrade. First, confirm the Enterprise Edition licence count matches the production CPU footprint after the upgrade — 23ai workloads frequently drive higher CPU consumption than equivalent 19c workloads, which can pull additional cores into scope. Second, confirm the supporting option entitlement (Partitioning, Advanced Compression, Active Data Guard) covers the planned vector workload deployment pattern. Third, confirm the Customer Service Identifier and the Order Form schedule covers the 23ai release explicitly.

The back-licence claim risk

The back-licence claim Oracle LMS deploys against AI Vector Search workloads is rarely against the VECTOR data type itself — it is against the supporting option footprint. The typical LMS finding pattern looks like this: "Customer enabled the VECTOR data type on a partitioned table without the Partitioning option licence active in the deployment", or "Customer deployed Active Data Guard for the vector workload without the ADG option licence on the standby database". The forensic defence is the option-by-option entitlement check before the workload deployment, not after the audit notice arrives. The Oracle audit guide sets out the LMS deployment patterns in detail.

The Autonomous Database 23ai economics

The ECPU consumption profile

Oracle Autonomous Database 23ai prices on a per-ECPU-per-hour consumption basis with the AI Vector Search functionality bundled at no incremental rate. The economic implication is that vector workloads on Autonomous land entirely in the ECPU consumption forecast. The forecast has to model three workload components: embedding generation throughput (CPU-intensive), vector index build operations (memory and CPU-intensive on initial deployment), and vector similarity query throughput (CPU and memory-intensive at query time). The aggregate ECPU consumption typically lands at 2× to 4× the equivalent transactional workload consumption.

The break-even versus self-managed Enterprise Edition

The break-even between Autonomous Database 23ai (consumption-based, no separate option licence) and self-managed Enterprise Edition 23ai (perpetual licence, separate option attach for production workload) depends on the workload profile and the existing licence footprint. For greenfield AI workloads without existing Oracle licence equity, Autonomous frequently wins on commercial simplicity. For brownfield workloads with material existing Enterprise Edition licence equity and supporting option entitlement, self-managed typically wins on total cost. The buyer-side analysis should model both paths against the actual workload forecast — not the account team's preferred narrative.

The OCI Universal Credits absorption

Autonomous Database 23ai consumption draws against the customer's OCI Universal Credits commitment. Customers with material Universal Credits commitments can negotiate the Autonomous ECPU rate against the commitment discount tier — typical commitment-based discount tiers run 25-45% off the published ECPU rate depending on the commitment size and the negotiation outcome. The negotiation should treat the Autonomous 23ai consumption as part of the broader Universal Credits commercial conversation. For the broader OCI commercial framework see the Oracle Cloud licensing guide.

"AI Vector Search is bundled with Oracle Database 23ai Enterprise Edition. The bundle is real. The buyer-side defence is to recognise that Oracle's playbook around bundled features routes the commercial exposure into the supporting option attach — Partitioning, Advanced Compression, Active Data Guard — and into the workload footprint expansion the AI workload drives across the licensed core count."

An anonymised case study — European retail bank, 23ai AI Vector Search deployment

A European retail bank with a 240-Processor Oracle Database Enterprise Edition footprint ran an AI Vector Search pilot in late 2025 to support a customer-service knowledge base retrieval pipeline. The original architecture proposed by the Oracle account team migrated the deployment to Autonomous Database 23ai with the AI Vector Search functionality bundled at the ECPU consumption rate. The projected annual consumption landed at $2.8m of ECPU-hour billing — a material incremental commitment against the existing Universal Credits position.

The buyer-side architecture review modelled three options. Option A was the proposed Autonomous Database 23ai migration at $2.8m projected annual consumption. Option B was a self-managed Enterprise Edition 23ai deployment using the existing 240-Processor entitlement, with incremental Partitioning, Advanced Compression, and Active Data Guard option licences for the vector workload — projected at $1.6m initial licence outlay plus $352k annual support uplift. Option C was a hybrid architecture — Autonomous 23ai for the variable production query traffic, self-managed Enterprise Edition for the predictable batch embedding pipeline, with the supporting option licences scoped to the self-managed footprint only.

The buyer-side recommendation was Option C with three commercial provisions. First, the Autonomous 23ai ECPU rate was renegotiated to absorb against the existing Universal Credits commitment at the 38% commitment discount tier, bringing the effective ECPU rate to $0.208 per ECPU-hour. Second, the supporting option licences on the self-managed footprint were procured at the negotiated 65% discount against list rather than the 30% the initial account team quote offered. Third, the workload deployment was scoped to a fixed core count with a written commercial provision that any expansion beyond the scope required commercial sign-off — protection against the workload creep that typically drives the back-licence claim. Net annualised cost across the architecture: $1.42m. Saving against the original Oracle proposal: $1.38m per year. For the broader Oracle commercial negotiation framework see our Oracle contract negotiation service.

The five buyer-side moves on Oracle 23ai AI Vector Search

Move 1 — Treat the bundle as the headline, not the answer. AI Vector Search is bundled with Enterprise Edition. The supporting option attach (Partitioning, Advanced Compression, Active Data Guard) is where the production deployment cost lands. Model the full option footprint before the upgrade business case.

Move 2 — Forensically size the workload footprint. Vector workloads typically drive 2× to 4× the CPU and memory consumption of equivalent transactional workloads. The 23ai upgrade may pull additional cores into the licensed footprint. Confirm the entitlement covers the post-upgrade CPU profile.

Move 3 — Model Autonomous against self-managed explicitly. The economic break-even depends on existing licence equity, the supporting option entitlement, and the workload variability. Model both paths against the actual workload forecast — not the account team's preferred narrative.

Move 4 — Negotiate the Universal Credits absorption on Autonomous consumption. Commitment-based discount tiering applies to Autonomous 23ai consumption. The negotiation should treat the AI workload as part of the broader Universal Credits commercial conversation, not as a separate transaction.

Move 5 — Defend against the back-licence claim with the entitlement check. The LMS finding pattern on AI workloads targets the supporting option footprint, not the VECTOR data type itself. Run the option-by-option entitlement check before the workload deployment, not after the audit notice arrives. Oracle 23ai AI Vector Search licensing rewards the customer who has done the forensic work upfront.

Frequently asked questions

Is Oracle 23ai AI Vector Search free?

Oracle Database 23ai AI Vector Search is bundled with the Oracle Database 23ai Enterprise Edition licence at no additional charge — it is not a separately licensed database option in the way that Advanced Compression, Partitioning, or RAC are separately licensed options. Customers running 23ai Enterprise Edition can enable and use the VECTOR data type, vector indexes, and vector similarity SQL operators without a separate option licence. However, AI Vector Search is not free in the absolute sense — the customer must hold a paid Oracle Database 23ai Enterprise Edition licence (Processor Metric or Named User Plus), and the storage and compute footprint required to host vector embeddings frequently drives an incremental licence position.

Can I use Oracle 23ai AI Vector Search on Standard Edition 2?

No. Oracle has restricted Oracle 23ai AI Vector Search to Enterprise Edition only. The VECTOR data type, vector indexes (HNSW, IVF), and the related SQL operators are not available on Standard Edition 2 in the 23ai release. Customers on SE2 wanting vector search functionality must either upgrade to Enterprise Edition (which is a significant commercial step) or use an external vector store (Pinecone, Weaviate, pgvector on PostgreSQL, or similar) and integrate at the application tier.

Does using Oracle 23ai AI Vector Search trigger an Oracle audit exposure?

AI Vector Search itself does not trigger a separate audit exposure because it is bundled with Enterprise Edition. However, the typical AI Vector Search workload triggers two distinct compliance gaps. First, vector embedding storage frequently drives an increase in database size that pulls the deployment toward additional database options (Partitioning for large vector tables, Advanced Compression for vector embedding compression). Second, the inference path for embedding generation frequently calls external services (OCI Generative AI Service for embeddings) which carries its own consumption-based billing against Universal Credits. Both vectors are typical examples of Oracle's playbook around AI workloads — the headline feature is bundled, the supporting workload footprint is where the exposure lands.

What is the licensing position for Oracle 23ai AI Vector Search on Autonomous Database?

Oracle Autonomous Database 23ai includes AI Vector Search at no additional charge — the consumption-based Autonomous Database pricing (per ECPU per hour) covers the AI Vector Search functionality without a separate uplift. The economic exposure on Autonomous shifts to the consumption profile — vector workloads typically drive higher CPU and storage consumption than transactional workloads, so the ECPU-hour and storage costs frequently increase materially when vector embeddings are deployed at scale.

Where 23ai Vector Search sits in the broader AI architecture

Storing embeddings inside Oracle Database 23ai with the bundled VECTOR datatype is only half the picture — the retrieval-augmented generation (RAG) pattern wires the vector index back into an LLM at query time, and that orchestration is where the licensing exposure compounds. Customers building RAG workloads on top of 23ai should benchmark the exposure forensic in our RAG architecture on Oracle Database licensing analysis, and audit the third-party framework integration patterns against our LangChain and LlamaIndex Oracle licensing risk analysis. The Oracle account team will frequently bundle 23ai Vector Search with the broader AI estate at renewal — the buyer-side defence is to right-size each component against actual workload, including the per-agent commercial mechanics in our Oracle AI Agents pricing analysis and the bundled-platform commercial pressure forensic in our Oracle AI Data Platform licensing analysis.

Related reading